Home / Current Issue / Paper 1722241
Hierarchical Prototype Network for Interpretable Chest X-ray Disease Classification
Subject area: Biological & Medical Sciences · Area of research: Medical Image Classification, Explainable AI
DOI: https://doi.org/10.64388/IREV10I2-1722241
Abstract
In recent years, deep learning has made great strides in medical image categorization, especially for automated chest X-ray image analysis for illness screening and early detection. Even though they are quite good at producing predictions, most deep learning-based models behave as black-box decision-making systems that don't give much information about how an input is classified. This could make clinician trust and interpretability weaker. To address this challenge, this research presents a Hierarchical Prototype network (HierProtoPnet) for interpretable chest X-ray classification of various images which are provided.nThe suggested solution uses a ResNet50 backbone for strong feature learning and two levels of prototype learning. This lets it learn both fine-grained and coarse-grained visual patterns in medical images at the same time. Lower-level prototypes learn more detailed local properties, such textures and edge patterns. Higher-level prototypes, on the other hand, encode more abstract disease aspects. Because of its hierarchical structure, the model classifies based on learnt prototypes, which gives clear predictions. The suggested method uses a ResNet50 backbone for strong feature extraction and has two levels of prototype learning. This allows it to learn both little and large visual patterns in medical images at the same time. Prototypes at a lower level learn more about the finer points of a place, like textures and edge patterns. Higher-level prototypes represent broader and more general disease characteristics. Because of this hierarchical structure, the model makes predictions by comparing input images with learned prototypes. This makes the reasoning behind each prediction clearer and easier to understand. The prototype-based approach also improves interpretability by showing how different visual patterns influence the final decision. The results indicate that hierarchical prototype learning supports diagnosis while improving both model accuracy and clarity, strengthening its suitability for clinical decision support systems. This study highlights the potential of combining deep convolutional feature extraction with hierarchical interpretability mechanisms to create accurate and informative models for medical image classification.
Keywords
Chest X-ray, Deep Learning, Hierarchical Prototype Network, ProtoPNet, ResNet50, Medical Image Classification, Explainable Artificial Intelligence (XAI)
How to cite this paper
@article{1722241,
author = {Akshaya Sreekumar, Sankalp Kumar Singh, Dr. Sibi Amaran, R. Shobana},
title = {Hierarchical Prototype Network for Interpretable Chest X-ray Disease Classification},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {10},
number = {2},
pages = {941-949},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1722241.pdf},
abstract = {In recent years, deep learning has made great strides in medical image categorization, especially for automated chest X-ray image analysis for illness screening and early detection. Even though they are quite good at producing predictions, most deep learning-based models behave as black-box decision-making systems that don't give much information about how an input is classified. This could make clinician trust and interpretability weaker. To address this challenge, this research presents a Hierarchical Prototype network (HierProtoPnet) for interpretable chest X-ray classification of various images which are provided.nThe suggested solution uses a ResNet50 backbone for strong feature learning and two levels of prototype learning. This lets it learn both fine-grained and coarse-grained visual patterns in medical images at the same time. Lower-level prototypes learn more detailed local properties, such textures and edge patterns. Higher-level prototypes, on the other hand, encode more abstract disease aspects. Because of its hierarchical structure, the model classifies based on learnt prototypes, which gives clear predictions.
The suggested method uses a ResNet50 backbone for strong feature extraction and has two levels of prototype learning. This allows it to learn both little and large visual patterns in medical images at the same time. Prototypes at a lower level learn more about the finer points of a place, like textures and edge patterns. Higher-level prototypes represent broader and more general disease characteristics. Because of this hierarchical structure, the model makes predictions by comparing input images with learned prototypes. This makes the reasoning behind each prediction clearer and easier to understand. The prototype-based approach also improves interpretability by showing how different visual patterns influence the final decision. The results indicate that hierarchical prototype learning supports diagnosis while improving both model accuracy and clarity, strengthening its suitability for clinical decision support systems. This study highlights the potential of combining deep convolutional feature extraction with hierarchical interpretability mechanisms to create accurate and informative models for medical image classification.},
keywords = {Chest X-ray, Deep Learning, Hierarchical Prototype Network, ProtoPNet, ResNet50, Medical Image Classification, Explainable Artificial Intelligence (XAI)},
month = {August},
doi = {https://doi.org/10.64388/IREV10I2-1722241}
}